hi.
I have been able to replicate the issue. It seems that the sampling frequency is not an integer as expected: sFile.prop.sfreq = 75.0019 whereas : mean(1./diff(nirs.t)) is 75.0002 (difference of 0.0017Hz)
So when creating the sample bounds; in the first case we try to access from 1 to 86407 ( round(sFile.prop.times(2) .* sFile.prop.sfreq) in the code) but the matrix data only contains 864006 samples. I think it is related to NIRS: Inconsistent sampling frequencies
An 'easy fix' is to only import the first 1150s of the signal as shown in the following screenshot 
Regards,
Edouard